Some speech recognition applications, like speaker verification, dialog recognition or the speech to text transcription could require real time processing and a good precision. Other applications such as toys, automotive vehicles or portable machines still could aggregate the portability and low-power requirements, in addition to physical compactness. These requirements could require a hardware solution for the speech recognition problem. The current work proposes an architecture using hardware based in FPGAs, optimizing the pre-processing and parameter extraction for performing efficient speech recognition.